Daily recap for June 3, 2026

Today we shipped 13 step-by-step tutorials for small-business AI — each one downloadable and ready to adapt. Among them, an auto-repair call intake agent that answers customer estimate calls while advisors stay in the bay, a per-tenant cost chargeback system for SaaS platforms, and a newsletter curator that replaces hours of manual work with an automated pipeline.

New tutorials

Auto-repair call intake agent for service advisors

Independent auto-repair shops lose bay time when advisors stop to answer estimate calls. This voice agent handles those calls — parsing the repair description, giving a price range, and only pulling in a human when needed. It connects Twilio, Deepgram, ElevenLabs, and OpenAI into a Next.js dashboard with full observability. Read the tutorial → Download the code (zip)

Built with agnostic voice-agent archetype, Next.js + Fastify, 119 tests, 100.00% coverage.

Per-tenant LLM cost tracking and chargeback for vertical SaaS

If your SaaS platform absorbs LLM costs across all tenants, margins shrink. This tutorial builds a per-tenant cost tracker that attributes every API call, computes real-time spend, and exports data for billing. It wires six REAA packages — SpendTracker, OTel bridge, cost telemetry, and multi-tenant accounting — into a Next.js application with a dashboard. Read the tutorial → Download the code (zip)

Built with agnostic cost-control archetype, Next.js + Fastify, 74 tests, 100.00% coverage.

Anthropic Knowledge Agent for Shopify Product Q&A

Shopify store owners get repetitive product questions. This recipe builds a knowledge agent that uses hybrid RAG (Qdrant + keyword search) and Anthropic Claude to answer questions about products, stock, and orders, with session memory and user preferences. It runs as a Next.js chat UI with three API endpoints. Read the tutorial → Download the code (zip)

Built with anthropic knowledge-agent archetype, Next.js, 111 tests, 99.64% coverage.

xAI Grok Observability for SMB AI Workflow Monitoring

If you use xAI Grok in your app, you need to watch token usage, latency, and errors. This observability tutorial gives you an instrumented Grok client that wraps every call with OpenTelemetry spans and cost telemetry, exporting to Langfuse. You’ll get a metrics dashboard and three API routes. Read the tutorial → Download the code (zip)

Built with xai-grok observability archetype, Next.js, 59 tests, 96.64% coverage.

Automated Industry Newsletter Curator

A small consulting firm’s weekly newsletter takes 5 hours of manual scanning and summarizing. This pipeline automates it: fetch RSS feeds, extract content, score relevance, generate AI summaries, and export a draft. It uses Hono, Readability, Qdrant, and any OpenAI-compatible LLM. Read the tutorial → Download the code (zip)

Built with agnostic document-pipeline archetype, Hono, 92 tests, 97.79% coverage.

xAI Grok Code Sandbox for Small Business Financial Modeling

Owners who need to run financial models without coding can describe them in plain English. This recipe uses Grok to generate and execute Python code in an e2b sandbox with strict safety policies, cost tracking, and spreadsheet export. Read the tutorial → Download the code (zip)

Built with xai-grok code-execution archetype, Express, 97 tests, 97.74% coverage.

Mistral AI Document Pipeline for NetSuite Invoice Automation

Finance teams waste hours keying invoice PDFs into NetSuite. This pipeline receives PDFs via webhook, extracts text with OCR, parses line items with Mistral AI, and posts draft vendor bills back to NetSuite, with deduplication and audit logging. Read the tutorial → Download the code (zip)

Built with mistral document-pipeline archetype, Next.js, 99 tests, 100.00% coverage.

Configurable PII redaction at ingress and egress for HIPAA, GDPR, and PCI compliance

Health-tech startups need to strip PHI before sending prompts to an LLM and catch any leaks in responses. This recipe builds dual-pipeline redaction for ingress and egress, configurable to HIPAA, GDPR, or PCI rules, with audit logging and zero external dependencies. Read the tutorial → Download the code (zip)

Built with agnostic security-guardrails archetype, Hono + Fastify, 125 tests, 98.59% coverage.

Review Response Agent for Independent Restaurants

Restaurant owners spend hours writing individual review replies. This agent generates consistent, on-brand responses across Yelp, Google, and TripAdvisor, using an LLM with response caching, guardrails, and memory to avoid repeating the same phrases. It exposes five API endpoints and a full test suite. Read the tutorial → Download the code (zip)

Built with agnostic knowledge-agent archetype, Next.js + Fastify, 108 tests, 95.37% coverage.

Automated Blog Content Pipeline with SEO Scoring

A SaaS company publishing two blog posts a week can cut the manual work. This pipeline takes a topic brief (title + keywords) and generates a post draft, then runs markdown validation, style linting, and SEO scoring — all via five API endpoints in Next.js. Read the tutorial → Download the code (zip)

Built with agnostic document-pipeline archetype, Next.js, 89 tests, 100.00% coverage.

Slack/email triage to priority inbox

Support teams drowning in Slack and email can let this agent triage. It reads incoming messages, classifies urgency, and drafts replies for the most critical issues, using intent classification, confidence-gated routing, and circuit breakers — all in a Next.js API. Read the tutorial → Download the code (zip)

Built with agnostic lead-intake archetype, Hono, 86 tests, 99.58% coverage.

Persistent customer memory across support sessions

Stop making customers repeat themselves. This recipe adds durable, per-customer memory to your support agent, storing conversation history, open tickets, and preferences, with a token budget that auto-compresses older messages. It uses Anthropic Claude and DynamoDB-backed session continuity. Read the tutorial → Download the code (zip)

Built with anthropic knowledge-agent archetype, Next.js + Fastify, 89 tests, 100.00% coverage.

Automated API key rotation for LLM providers with zero downtime

Manual API key rotation causes outages. This recipe builds a secret rotation sidecar that swaps keys for OpenAI, Anthropic, etc., across AWS Secrets Manager or Vercel, without interrupting running agents. You get a CLI tool and an HTTP server with status and validation endpoints. Read the tutorial → Download the code (zip)

Built with agnostic reliability-ops archetype, CLI + Hono, 125 tests, 97.50% coverage.

Building blocks shipped

agent-auth-proxy

An identity-aware proxy for agent-to-service communication, with schemas, a typed client SDK, and a Fastify server plugin. The core package provides shared validations, the client exchanges API keys for JWTs, and the server enforces OAuth2 scopes. Browse the agent-auth-proxy family →

agent-budget-controller

A complete budget enforcement kit for agents: real-time spend tracking, pricing engine, pre-flight cost checks, and automatic model downgrade. Standout packages include the budget engine that enforces policies, middleware for request interception, and a CLI for management. Browse the agent-budget-controller family →

circuit-breaker-agents

Circuit breaker state machines for tool and agent communication, with pluggable persistence (Redis, DynamoDB, Firestore). The core package handles trip strategies and recovery, while persistence adapters keep state across restarts and instances. Browse the circuit-breaker-agents family →

llm-router

A config-driven LLM routing engine with model selection, fallback chains, and cost tracking, plus a CLI for benchmarking. The engine orchestrates decisions; the CLI helps test and report. Browse the llm-router family →

webhook-relay-mcp

A full MCP server for ingesting and relaying webhooks from Stripe, GitHub, Twilio, and others into agent workflows. It ships with SQLite storage, 15 management tools, and signature validation. The core types, storage layer, and CLI binary are the key pieces. Browse the webhook-relay-mcp family →

mcp-server-starter-ts

An Express-based MCP server framework with tool registration, dual transport (HTTP + SSE), auth, rate limiting, and observability. The engine handles orchestration, tools package simplifies tool creation, and observability adds tracing and metrics. Browse the mcp-server-starter-ts family →

confidence-router

A threshold-based decision engine that routes, clarifies, or falls back based on classification confidence. Includes pluggable classifiers (keyword, embedding, LLM) and a grid-search optimizer for thresholds. Browse the confidence-router family →

media-pipeline-mcp

An MCP server exposing 35+ media operations (image, audio, video) with provider routing and cost tracking. The server bundles all, and separate provider packages (Deepgram, Replicate, Google) handle specific tasks. Browse the media-pipeline-mcp family →

classifier-evals

A comprehensive evaluation toolkit for classifiers: dataset loading, 14 metrics, regression gates, LLM judge, and exporters. The metrics engine gives you accuracy, F1, and more; gates enforce thresholds automatically. Browse the classifier-evals family →

mcp-contract-kit

Conformance testing for MCP servers, with validators for protocol, registry, routing, security, and performance. The CLI runs tests, validators check compliance, and reporters output results in multiple formats. Browse the mcp-contract-kit family →

Browse the full catalog at reaatech.com/products.